Research On Application Of Surfacelet Transform In Video Processing

The rapid development and wide application of multimedia technology allow the video processing technology has attracted great attention, video denoising and dynamic texture retrieval are hot issues in the research of video processing.Surfacelet transform is a processing tool in high-dimensional signal with characteristics of multi-directional, anisotropic and low redundancy, it is the combination of the multi-scale pyramid and the multi-dimensional directional filter banks. In this paper, the applications of the Surfacelet transform in dynamic texture retrieval and video denoising are mainly studied, The major works can be summarized as follows:(1) A dynamic texture retrieval method based on the Surfacelet transform is proposed. Based on the distribution characteristics of Surfacelet coefficients of dynamic texture, according the Shannon entropy of the energy sequence of the Surfacelet directional subband, we classify the dynamic texture as structural dynamic texture and random dynamic texture. The Parameters of GGD and energy of every directional subband in the Surfacelet transform domain are used as the features of dynamic texture, and the Euclidean distance and Kullbaek-Leibler distance are used as the similarity measurements for the dynamic texture.(2) An adaptive video denoising method based on the Surfacelet transform is proposed. In consideration of the characteristics of the distribution of the Surfacelet coefficients of video and noise, based on the neighborhood relations of Surfacelet coefficients, the characteristics of the energy of video and noise after Surfacelet transform are analyzed. And the adaptive video denoising method is derived.(3) A video denoising method based on the BKF distribution of the Surfacelet coefficients and Bayesian is proposed. The method model Surfacelet coefficients with the BKF distribution, combining Bayesian theory, the relations of Surfacelet coefficients are made full used. The noises are well denoised with the presented algorithm.This work is supported by National Natural Science Foundation of China (No.60972148) and Special Fund of Central University Basic Research And Operating Expenses (No. JY10000902043).